tractable learning in structured probability spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna...
TRANSCRIPT
![Page 1: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/1.jpg)
Tractable Learning in Structured Probability Spaces
Adnan Darwiche Computer Science Department
Representation learning workshop Simons Institute, 3/29/2017
![Page 2: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/2.jpg)
Probabilistic Sentential Decision Diagrams Doga Kisa, Guy Van den Broeck, Arthur Choi and Adnan Darwiche KR, 2014
Learning with Massive Logical Constraints Doga Kisa, Guy Van den Broeck, Arthur Choi and Adnan Darwiche ICML 2014 workshop
Tractable Learning for Structured Probability Spaces Arthur Choi, Guy Van den Broeck and Adnan Darwiche IJCAI, 2015
Tractable Learning for Complex Probability Queries Jessa Bekker, Jesse Davis, Arthur Choi, Adnan Darwiche, Guy Van den Broeck. NIPS, 2015
Structured Features in Naive Bayes Classifiers Arthur Choi, Nazgol Tavabi and Adnan Darwiche AAAI, 2016
Tractable Operations on Arithmetic Circuits Jason Shen, Arthur Choi and Adnan Darwiche NIPS, 2016
References
![Page 3: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/3.jpg)
Structured probability spaces?
![Page 4: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/4.jpg)
Courses: • Logic (L) • Knowledge Representation (K) • Probability (P) • Artificial Intelligence (A)
Data
• Must take at least one of Probability or Logic.
• Probability is a prerequisite for AI. • The prerequisites for KR is
either AI or Logic.
Prior Knowledge
Running Example
![Page 5: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/5.jpg)
L K P A0 0 0 00 0 0 10 0 1 00 0 1 10 1 0 00 1 0 10 1 1 00 1 1 11 0 0 01 0 0 11 0 1 01 0 1 11 1 0 01 1 0 11 1 1 01 1 1 1
unstructured
Probability Space
![Page 6: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/6.jpg)
L K P A0 0 0 00 0 0 10 0 1 00 0 1 10 1 0 00 1 0 10 1 1 00 1 1 11 0 0 01 0 0 11 0 1 01 0 1 11 1 0 01 1 0 11 1 1 01 1 1 1
unstructured L K P A0 0 0 00 0 0 10 0 1 00 0 1 10 1 0 00 1 0 10 1 1 00 1 1 11 0 0 01 0 0 11 0 1 01 0 1 11 1 0 01 1 0 11 1 1 01 1 1 1
structured
Structured Probability Space
7 out of 16 instantiations are impossible
• Must take at least one of Probability or Logic.
• Probability is a prerequisite for AI. • The prerequisites for KR is
either AI or Logic.
![Page 7: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/7.jpg)
Learning with Constraints
Learn a statistical model that assigns zero probability to instantiations that violate the constraints.
Data
Constraints (Background Knowledge) (Physics)
Statistical Model (Distribution) Learn
![Page 8: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/8.jpg)
Example: Video
[Lu, W. L., Ting, J. A., Little, J. J., & Murphy, K. P. (2013). Learning to track and identify players from broadcast sports videos.]
![Page 9: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/9.jpg)
Example: Language
• Non-local dependencies: At least one verb in each sentence
• Sentence compression If a modifier is kept, its subject is also kept
• Information extraction • Semantic role labeling • … and many more!
[Chang, M., Ratinov, L., & Roth, D. (2008). Constraints as prior knowledge],…, [Chang, M. W., Ratinov, L., & Roth, D. (2012). Structured learning with constrained conditional models.], [https://en.wikipedia.org/wiki/Constrained_conditional_model]
![Page 10: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/10.jpg)
Bayesian network synthesized from specs of power system (NASA Ames):
Has many constraints (0/1 parameters) due to domain ``physics’’
![Page 11: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/11.jpg)
Example: Deep Learning
[Graves, A., Wayne, G., Reynolds, M., Harley, T., Danihelka, I., Grabska-Barwińska, A., et al.. (2016). Hybrid computing using a neural network with dynamic external memory. Nature, 538(7626), 471-476.]
![Page 12: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/12.jpg)
What are people doing now?
• Ignore constraints • Handcraft into models • Use specialized distributions • Find non-structured encoding • Try to learn constraints • Hack your way around
Accuracy ? Specialized skill ?
Intractable inference ? Intractable learning ? Waste parameters ?
Risk predicting out of space ?
you are on your own
![Page 13: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/13.jpg)
Structured Probability Spaces
• Everywhere in ML! – Configuration problems, inventory, video, text, deep learning – Planning and diagnosis (physics) – Causal models: cooking scenarios (interpreting videos) – Combinatorial objects: parse trees, rankings, directed acyclic graphs,
trees, simple paths, game traces, etc.
No statistical ML boxes out there
that take constraints as input!
Goal: Constraints as important as data! General purpose!
![Page 14: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/14.jpg)
Specification Language: Logic
![Page 15: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/15.jpg)
L K P A0 0 0 00 0 0 10 0 1 00 0 1 10 1 0 00 1 0 10 1 1 00 1 1 11 0 0 01 0 0 11 0 1 01 0 1 11 1 0 01 1 0 11 1 1 01 1 1 1
unstructured L K P A0 0 0 00 0 0 10 0 1 00 0 1 10 1 0 00 1 0 10 1 1 00 1 1 11 0 0 01 0 0 11 0 1 01 0 1 11 1 0 01 1 0 11 1 1 01 1 1 1
structured
Structured Probability Space
7 out of 16 instantiations are impossible
• Must take at least one of Probability or Logic.
• Probability is a prerequisite for AI. • The prerequisites for KR is
either AI or Logic.
![Page 16: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/16.jpg)
L K P A0 0 0 00 0 0 10 0 1 00 0 1 10 1 0 00 1 0 10 1 1 00 1 1 11 0 0 01 0 0 11 0 1 01 0 1 11 1 0 01 1 0 11 1 1 01 1 1 1
unstructured L K P A0 0 0 00 0 0 10 0 1 00 0 1 10 1 0 00 1 0 10 1 1 00 1 1 11 0 0 01 0 0 11 0 1 01 0 1 11 1 0 01 1 0 11 1 1 01 1 1 1
structured
Boolean Constraints
7 out of 16 instantiations are impossible
![Page 17: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/17.jpg)
Combinatorial Objects: Rankings
10 items: 3,628,800 rankings
rank sushi
1 fatty tuna
2 sea urchin
3 salmon roe
4 shrimp
5 tuna
6 squid
7 tuna roll
8 see eel
9 egg
10 cucumber roll
rank sushi
1 shrimp
2 sea urchin
3 salmon roe
4 fatty tuna
5 tuna
6 squid
7 tuna roll
8 see eel
9 egg
10 cucumber roll
20 items: 2,432,902,008,176,640,000 rankings
![Page 18: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/18.jpg)
Combinatorial Objects: Rankings
rank sushi
1 fatty tuna
2 sea urchin
3 salmon roe
4 shrimp
5 tuna
6 squid
7 tuna roll
8 see eel
9 egg
10 cucumber roll
rank sushi
1 shrimp
2 sea urchin
3 salmon roe
4 fatty tuna
5 tuna
6 squid
7 tuna roll
8 see eel
9 egg
10 cucumber roll
Aij item i at position j (n items require n2 Boolean variables)
An item may be assigned to more than one position
A position may contain more than one item
![Page 19: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/19.jpg)
Encoding Rankings in Logic
Aij : item i at position j pos1 pos2 pos3 pos4
item1 A11 A12 A13 A14item2 A21 A22 A23 A24item3 A31 A32 A33 A34item4 A41 A42 A43 A44
constraint: each item i assigned to a unique position (n constraints)
constraint: each position j assigned a unique item (n constraints)
![Page 20: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/20.jpg)
Unstructured probability space: 184+16,777,032 = 224
Structured Space for Paths
Good variable assignment(represents route)
184
Bad variable assignment(does not represent route)
16,777,032
Space easily encoded in logical constraints
![Page 21: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/21.jpg)
the
DT
cat
NN
NP
sleeps
Vi
VP
S
dog
NN
NP
saw
Vt
VP
S
the
DT
the
DT
cat
NN
NP
Parse Trees
Undirected Graphs (Unstructured)
Trees
Labeled Treesdog
cat
dog
S S
VP
VP
S
S
S
S
Acyclicity Constraints
Label Constraints (CFG Production Rules)
![Page 22: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/22.jpg)
“Deep Representation”
Logic + Probability
![Page 23: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/23.jpg)
¬L K L ⊥ P A ¬P ⊥ L ¬L ⊥ ¬P ¬A P ¬L ¬K L ⊥ P ¬P ⊥
K ¬K A ¬A A ¬A
Logical Circuits
![Page 24: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/24.jpg)
¬L K L ⊥ P A ¬P ⊥ L ¬L ⊥ ¬P ¬A P ¬L ¬K L ⊥ P ¬P ⊥
K ¬K A ¬A A ¬A
Property: Decomposability
![Page 25: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/25.jpg)
¬L K L ⊥ P A ¬P ⊥ L ¬L ⊥ ¬P ¬A P ¬L ¬K L ⊥ P ¬P ⊥
K ¬K A ¬A A ¬A Input: L, K, P, A
Property: Determinism
![Page 26: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/26.jpg)
¬L K L ⊥ P A ¬P ⊥ L ¬L ⊥ ¬P ¬A P ¬L ¬K L ⊥ P ¬P ⊥
K ¬K A ¬A A ¬A Input: L, K, P, A
Sentential Decision Diagram (SDD)
![Page 27: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/27.jpg)
Tractable for Logical Inference
• Is structured space empty? (SAT) • Count size of structured space (#SAT) • Check equivalence of spaces • Algorithms linear in circuit size
(pass up, pass down, similar to backprop)
![Page 28: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/28.jpg)
¬L K L ⊥
1 0
P A ¬P ⊥
1 0
L ¬L ⊥
1 0
¬P ¬A P
0.6 0.4
¬L ¬K L ⊥
1 0
P ¬P ⊥
1 0
K ¬K 0.8 0.2
A ¬A 0.25 0.75
A ¬A 0.9 0.1
0.1 0.6 0.3
PSDD: Probabilistic SDD
![Page 29: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/29.jpg)
¬L K L ⊥
1 0
P A ¬P ⊥
1 0
L ¬L ⊥
0
¬P ¬A P
0.6
¬L ¬K L ⊥
1 0
P ¬P ⊥
1 0
K ¬K 0.2
A ¬A 0.75
A ¬A 0.9 0.1
0.1 0.6
Input: L, K, P, A
PSDD: Probabilistic SDD
1
0.3
0.4
0.8 0.25
![Page 30: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/30.jpg)
¬L K L ⊥
1 0
P A ¬P ⊥
1 0
L ¬L ⊥
1.0 0
¬P ¬A P
0.6 0.4
¬L ¬K L ⊥
1 0
P ¬P ⊥
1 0
K ¬K 0.8 0.2
A ¬A 0.25 0.75
A ¬A 0.9 0.1
0.1 0.6 0.3
Input: L, K, P, A Pr(L,K,P,A) = 0.3 x 1.0 x 0.8 x 0.4 x 0.25 = 0.024
PSDD: Probabilistic SDD
![Page 31: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/31.jpg)
¬L K L ⊥
1 0
P A ¬P ⊥
1 0
L ¬L ⊥
1 0
¬P ¬A P
0.6 0.4
¬L ¬K L ⊥
1 0
P ¬P ⊥
1 0
A ¬A 0.8 0.2
A ¬A 0.25 0.75
A ¬A 0.9 0.1
0.1 0.6 0.3
Can read probabilistic independences off the circuit structure
PSDD nodes induce a normalized distribution!
![Page 32: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/32.jpg)
Tractable for Probabilistic Inference
• MAP inference: Find most-likely assignment (otherwise NP-complete)
• Computing conditional probabilities Pr(x|y) (otherwise PP-complete)
• Sample from Pr(x|y)
• Algorithms linear in circuit size (pass up, pass down, similar to backprop)
![Page 33: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/33.jpg)
Known in the ML literature as SPNs UAI 2011, NIPS 2012 best paper awards
PSDDs are Arithmetic Circuits [Darwiche, JACM 2003]
[ICML 2014] (SPNs equivalent to ACs)
α2
α1 αn
p1 s1 p2 s2 pn sn
PSDD AC
+
* * *
* * * α1 α2 αn
p1 s1 p2 s2 pn sn
![Page 34: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/34.jpg)
Learning PSDDs
Logic + Probability + ML
![Page 35: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/35.jpg)
¬L K L ⊥
1 0
P A ¬P ⊥
1 0
L ¬L ⊥
1 0
¬P ¬A P
0.6 0.4
¬L ¬K L ⊥
1 0
P ¬P ⊥
1 0
K ¬K 0.8 0.2
A ¬A 0.25 0.75
A ¬A 0.9 0.1
0.1 0.6 0.3
Student takes course L
Student takes course P
Probability of P given L
Parameters are Interpretable
Explainable AI DARPA Program
![Page 36: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/36.jpg)
Learning Algorithms
• Parameter learning: Closed form max likelihood from complete data One pass over data to estimate parameters
• Structure learning: – Compile constraints to SDD (naive)
Use SAT solver technology (SDD library)
– Search for structure to fit data (ongoing work)
Note a lot to say: very easy!
![Page 37: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/37.jpg)
Learning Preference Distributions
Special-purpose distribution: Mixture-of-Mallows
– # of components from 1 to 20
– EM with 10 random seeds
– implementation of Lu & Boutilier
PSDD
This is the naive approach, without real structure learning!
![Page 38: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/38.jpg)
What happens if you ignore constraints?
![Page 39: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/39.jpg)
X X OO O XX X O
X O XO XX O
O OX O
X X X
optimal, heuristic, random
Attribute with 362,880 values (possible game traces)
Structured Naïve Bayes Classifier
X1 X2 Xn
C
…
![Page 40: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/40.jpg)
s
t
s
t
s
t
X1 X2 Xn
C
…
normal, abnormal
Attribute with 789,360,053,252 values (routes in 8 × 8 grid) Ongoing work: learn anomalies from Uber data
Structured Naïve Bayes Classifier
![Page 41: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/41.jpg)
Structured datasets and queries
![Page 42: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/42.jpg)
Incomplete Data
id X Y Z
1 x1 y2 z1
2 x2 y1 z2
3 x2 y1 z2
4 x1 y1 z1
5 x1 y2 z2
a classical complete dataset
id X Y Z
1 x1 y2 ? 2 x2 y1 ? 3 ? ? z2
4 ? y1 z1
5 x1 y2 z2
a classical incomplete dataset
a new type of incomplete dataset
id X Y Z
1 X ≡ Z
2 x2 and (y2 or z2)
3 x2 ⇒ y1
4 X ⊕ Y ⊕ Z ≡ 1
5 x1 and y2 and z2
closed-form (maximum-likelihood estimates are unique)
EM algorithm (on PSDDs)
Missed in the ML literature
![Page 43: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/43.jpg)
id 1st sushi
2nd sushi
3rd sushi …
1 fatty tuna
sea urchin
salmon roe …
2 fatty tuna tuna shrimp …
3 tuna tuna roll
sea eel …
4 fatty tuna
salmon roe tuna …
5 egg squid shrimp …
a classical complete dataset (e.g., total rankings)
id 1st sushi
2nd sushi
3rd sushi …
1 fatty tuna
sea urchin ? …
2 fatty tuna ? ? …
3 tuna tuna roll ? …
4 fatty tuna
salmon roe ? …
5 egg ? ? …
a classical incomplete dataset (e.g., top-k rankings)
Structured Datasets
![Page 44: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/44.jpg)
id 1st sushi
2nd sushi
3rd sushi …
1 fatty tuna
sea urchin
salmon roe …
2 fatty tuna tuna shrimp …
3 tuna tuna roll
sea eel …
4 fatty tuna
salmon roe tuna …
5 egg squid shrimp …
a classical complete dataset (e.g., total rankings)
id 1st sushi
2nd sushi
3rd sushi …
1 (fatty tuna > sea urchin) and (tuna > sea eel) …
2 (fatty tuna is 1st) and (salmon roe > egg) …
3 tuna > squid …
4 egg is last …
5 egg > squid > shrimp …
a new type of incomplete dataset (e.g., partial rankings)
(represents constraints on possible total rankings)
Structured Datasets
![Page 45: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/45.jpg)
Learning from Incomplete Data
• Movielens Dataset: – 3,900 movies, 6,040 users, 1m ratings
– take ratings from 64 most rated movies
– ratings 1-5 converted to pairwise prefs.
• PSDD for partial rankings – 4 tiers – 18,711 parameters
rank movie
1 The Godfather
2 The Usual Suspects
3 Casablanca
4 The Shawshank Redemption
5 Schindler’s List
6 One Flew Over the Cuckoo’s Nest
7 The Godfather: Part II
8 Monty Python and the Holy Grail
9 Raiders of the Lost Ark
10 Star Wars IV: A New Hope
movies by expected tier
![Page 46: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/46.jpg)
PSDD Sizes
![Page 47: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/47.jpg)
Structured Queries
rank movie
1 Star Wars V: The Empire Strikes Back
2 Star Wars IV: A New Hope
3 The Godfather
4 The Shawshank Redemption
5 The Usual Suspects
• no other Star Wars movie in top-5 • at least one comedy in top-5
rank movie
1 Star Wars V: The Empire Strikes Back
2 American Beauty
3 The Godfather
4 The Usual Suspects
5 The Shawshank Redemption
diversified recommendations via logical constraints
![Page 48: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/48.jpg)
Conclusions • Structured spaces are everywhere • Roles of Boolean constraints in ML
– Domain constraints and combinatorial objects (structured probability space)
– Incomplete examples (structured datasets)
– Questions and evidence (structured queries)
• Learn distributions over combinatorial objects
• Strong properties for inference and learning: Probabilistic sentential decision diagram (PSDD)
![Page 49: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/49.jpg)
Conclusions
Statistical ML “Probability”
Symbolic AI “Logic”
Connectionism “Deep”
PSDD
![Page 50: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/50.jpg)
Probabilistic Sentential Decision Diagrams Doga Kisa, Guy Van den Broeck, Arthur Choi and Adnan Darwiche KR, 2014
Learning with Massive Logical Constraints Doga Kisa, Guy Van den Broeck, Arthur Choi and Adnan Darwiche ICML 2014 workshop
Tractable Learning for Structured Probability Spaces Arthur Choi, Guy Van den Broeck and Adnan Darwiche IJCAI, 2015
Tractable Learning for Complex Probability Queries Jessa Bekker, Jesse Davis, Arthur Choi, Adnan Darwiche, Guy Van den Broeck. NIPS, 2015
Structured Features in Naive Bayes Classifiers Arthur Choi, Nazgol Tavabi and Adnan Darwiche AAAI, 2016
Tractable Operations on Arithmetic Circuits Jason Shen, Arthur Choi and Adnan Darwiche NIPS, 2016
References
![Page 51: Tractable Learning in Structured Probability Spaces€¦ · 29/03/2017 · rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d472d5824a2a5658b6a5/html5/thumbnails/51.jpg)
Questions?
PSDD